The performance of model-based battery algorithms depends on the fidelity of the adopted battery model including the open-circuit voltage (OCV) versus state-of-charge (SoC) relationship. The OCV-SoC model is usually extracted using offline experiments at the beginning-of-life (BoL) and is often assumed not to change over the battery lifetime. However, as shown in recent literature and this article, the battery OCV-SoC curve will change because of cyclic and calendar aging. Thus, the effect of aging on the OCV-SoC curve should be compensated to maintain accuracy in different battery algorithms. To address this gap, this article puts forward a reinforcement learning (RL) framework to train an artificial intelligence agent that can adjust and compensate for the changes of the OCV-SoC curve in different temperature and aging conditions. First, parameter and state estimators are designed to estimate the OCV and SoC terms and the feedback is then provided to train the RL agent used for prediction of the parameters of the OCV curve model. The effectiveness of the proposed method is demonstrated in SoC estimation using actual battery performance and lifetime data related to four high-capacity lithium-ion cells. The proposed RL agent yields an average RMSE of about 10 mV and improves SoC estimation accuracy up to 7% when the agent is applied to compensate OCV-SoC curve of the aged cells.

In this paper, an analysis and performance review of a unique hybrid high-power lithium-iron phosphate cell (HP-LFP) with a high cycle life and fast charge/discharge rate is presented. The new hybrid cell has been developed under the framework of the EU-funded project Hybrid Energy Storage Station (HEROES). The proposed HP-LFP cell simultaneously achieves an optimized energy and power throughput making it useful for applications where both power and energy are demanded such as different types of electric vehicles (EVs), microgrids, and charge stations. The cathode of the hybrid HP-LFP cell is similar to the LFP batteries while the supercapacitorlike anode is based on hard carbon. Thus, the HP-LFP cell has some similarities to the recent lithium-ion capacitors (LiCs) but it behaves differently than the existing LFP batteries and supercapacitors. To distinguish the differences, HP-LFP cell samples are tested, characterized, and analyzed to review their performance compared to the existing technologies. Different standard tests including the open-circuit test, hybrid pulse power characteristics test, and capacity test are fulfilled. The test data is used for model parameterization and parameter sensitivity analysis with respect to the cell's state of charge, temperature, and charge rate. The cell performance has been validated through mission profile testing in an EV fast charge station. Based on the analysis, the paper provides a comparison between different energy storage cells.

State estimation based on Kalman Filter (KF) has been widely considered for state-of-charge (SoC) prediction of lithium-ion (Li-ion) batteries. However, the KF is a model-based approach, and its performance declines when faced with modeling inaccuracies resulting from the nonlinear and timevarying nature of Li-ion batteries. To tackle the modeling uncertainties, a novel hybrid variant of KF is deployed, in which priori estimations are attained similar to the KF algorithm while the posterior estimations are obtained using a recurrent neural network (RNN) integrated into the KF architecture. The proposed method yields enhanced robustness in maintaining the accuracy of SoC estimations in the presence of model mismatches, e.g. due to the aging of batteries. In the proposed method, named KalmanNet, the RNN learns from battery data to predict the optimum Kalman gains for the minimization of SoC estimation error. The proposed method is trained and tested using actual battery data of a high-capacity pouch Li-ion cell replicating real-life driving cycles of electric vehicles. The significance of the results is twofold: 1-When exposed to a model mismatch of ±5%, the SoC estimation accuracy is improved by about 0.5% compared to EKF. 2-The size of required training data is reduced by 20% compared to similar machine learning algorithms.